Estimating the soil content using near infrared (NIR) spectroscopy with an appropriate method of multivariate regression analysis, is a rapid and nondestructive testing technique with the virtue of high analysis speed and easy operation. Partial least squares (PLS) and least square support vector machine (LS-SVM) as the existing models widely used in other studies were improved and employed to develop an optimal regression model for the prediction of typical soil nutrient in Wuling mountain, Hunan Province, including total nitrogen, available phosphorus, and organic carbon. The performance of models established in this research was assessed by the coefficient of determination (R) and the root mean square error of calibration (RMSEC) and prediction (RMSEP). The result showed that the pre-processing method MSC SG displayed the highest R values in PLS and LS-SVM models, which were 0.89 and 0.91, respectively. However, compared to the PLS model, LS-SVM displayed more desirable performance on the predictions. The RSMEC and RMSEP values of LS-SVM (4.84 and 4.75, respectively) were much better compared to PLS (6.15 and 6.58, respectively).
展开▼
